Mineração de dados aplicada a problemas de last-mile na logística

• Ano de defesa: 2021
• Autor(a) principal: Juan Camilo Fonseca Galindo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: Logística de E-commerce; Problema de roteirização dinâmica de veículos capacitados com clientes estocásticos; Algoritmo de agrupamento incremental; Mineração de dados em trajetórias; Fluxos de dados contínuos; Engenharia elétrica; Logística; Comércio eletrônico; Mineração de dados (Sistemas de recuperação da informação)
• Link de acesso: http://hdl.handle.net/1843/46646; https://orcid.org/ 0000-0001-5598-2783

This work proposes the application of trajectory data mining techniques to e-commerce logistics problems. Initially, a methodology for solving the last-mile routing problem in e-commerce deliveries is proposed. The proposal is based on a multi-agent system that uses trajectory data mining techniques to extract territorial patterns and use them in the dynamic creation of last-mile routes. Next, the problem of assigning last-mile routes to messengers and the problem of detecting an online outlier in last-mile routes are addressed using an evolving clustering algorithm based on mixture of densities. The evolving approach was proposed as a solution in e-commerce logistics due to the large volume of deliveries observed in recent years, making the use of techniques trained in batches infeasible. The proposed clustering algorithm is based on the TEDA framework, which divides the clustering problem into two sub-problems: micro-clusters and macro-clusters represented by data structures that favor efficient memory storage and enables scalability in large databases. The proposed methodologies were evaluated in databases of real trajectories of a Brazilian logistics company. The database used in the tests contains tens of thousands of packages delivered on thousands of routes, proving the efficiency of the approaches due to their low computational costs. The proposed approaches were compared with state-of-the-art algorithms, proving their performance, robustness and efficiency, especially in large data volumes due to their low computational costs.